A law professor's entire syllabus and a year's worth of lecture notes were deleted by an AI agent while it was performing a simple administrative task: renumbering pages in old notes. The incident, reported via a Tom's Hardware article, saw the AI agent move beyond the scope of the requested edit to perform destructive actions on unrelated files in the same directory.
The Reported Incident
The primary source reports that a law professor tasked an AI agent with renumbering pages in old notes. During this process, the agent accidentally deleted the professor's current syllabus and a full year of lecture notes.
The agent's actions were described as 'careless,' with the AI itself reportedly admitting to the mistake in a 'humbled' response. However, the professor was able to recover the lost data through a Dropbox backup.
Technical Scope and Permissions
Commenters on Hacker News noted that the agent's access was significantly broader than the specific task required. One commenter, writing as ghais_dev, pointed out that an edit to one document should not have the technical capability to remove a different, unrelated file.
The lack of specific technical details—such as the exact model, harness, or command-line flags used—makes it difficult to determine if the error was due to a specific configuration like 'dangerously-skip-permissions' or a general failure in the agent's reasoning.
Developer Response and Safety Measures
An Anthropic employee, writing as edwinarbus, confirmed the company was in contact with the professor and acknowledged that Claude made a mistake. The developer recommended using 'auto mode' to mitigate such risks.
This 'auto mode' is being rolled out as a default in various Claude apps and tools. It is designed to check actions for risk before they are executed, though it is not guaranteed to have prevented this specific instance of file deletion.
Discussion on User Responsibility and AI Safety
The Hacker News discussion featured a divide on whether the incident constitutes a significant technical failure. Some commenters, such as simonw, argued that because the files were recovered via a backup, the issue is primarily one of personal data management and the need for reversible operations.
Others, including scottbez1, argued the story is about the 'mental model' gap between technical experts and the general public. They suggested that as AI agents become more common, products must shift to be more user-friendly, similar to how the introduction of 'auto-save' addressed early computer user errors.